If I was a Betting Man

Let me first state that I do not gamble.
My endeavor is to provide a free service, without any strings attached, motivated purely by curiosity. This project is still in its infancy, so all results should be taken with a grain of salt.

So let's build a model. To narrow the scope, I will be only concentrating on OVER/UNDER wagering. We begin by laying down some assumptions and observations:

• The bookmakers want a LINE that has about the same money on both sides. This guarantees that they will make money on commissions, regardless of the outcome. Thus, the resulting LINE may or may not be statistically indifferent. What I mean by that is simple: the line doesn't necessarily give you a 50/50 chance of winning. The biases of all the individuals of the betting pool factor into where the LINE moves.
• This model attempts to find the true LINE, the one free of biases. After which if it is found to be statistically significantly different from the biased line, then the appropriate betting recommendation is given.
• The model will only make a bet if it is favored to do so; therefore, it will not make a bet just for the sake of betting.
• The OVER/UNDER LINE represents the predicted sum of two teams. Therefore, points and points allowed should be analyzed.
• Teams generally play better at home than on the road. Thus, home scoring for the home team and road scoring for the away team should also be analyzed.
• The only data that are processed are the end scores. This is done so to maintain simplicity as there are infinite potential conditions that could be considered.
• Assuming a juice rate of 10%, the model has to be right at least 52.38% of the time. This also assumes that the same sized wager is made for the long run.
• This model will most likely have "bad streaks," just like how Shaq might have missed five free throws in a row. But even his career FT% of 52.7 would make money in the long run.

OK, now the structure of the model is laid down. It's time to run through how the model operates:

1. Input the AWAY and HOME teams
2. Model gathers the scoring average [SA], opponent scoring average [OSA], and road scoring average [RSA] for the AWAY team
3. Model also gathers the SA, OSA, and home scoring average [HSA] for the HOME team
4. Input the betting LINE
5. Model factors in skew characteristics for all relevant variables (an explanation for this step would require its own thread). Adjust variables from step 3 if needed.
6. Model compares AWAY SA + HOME SA, AWAY OSA + HOME OSA, RSA + HSA to LINE
7. Model provides three recommendations based on the three comparisons in step 6
8. If the three recommendations from step 7 are the same, then the appropriate OVER/UNDER bet will be recommended. If they aren't the same, then NO BET is recommended.

Let's run through an example:

1. We'll set the risk tolerance (actually risk aversion rate) to 53%. Each recommendation has to be at least 53% statistically sure for OVER/UNDER, otherwise, NO BET
2. For the teams, we'll use LAL/DEN on FEB 25, 2013
3. AWAY SA + HOME SA = 207.7; 210.7(skew adjusted)
4. AWAY OSA + HOME OSA = 203; 203
5. RSA + HSA = 211.1; 211.6
6. LINE of 217.5
7. First recommendations: UNDER 68.3%/UNDER 80.6%/UNDER 67.5%
8. Last recommendations: UNDER because previous recommendations matched with reasonable certainty
9. Watch the game
10. Curse obscenities as LAL/DEN goes OVER

I posted in chunks, because it kept giving me a 7800 character limit error (even though my original post was under).

---

Previous Picks: (5-3)

FEB 24, 2013

• LAL/DAL   UNDER   WIN
• GS/MIN   UNDER   WIN
• CLE/MIA   NO BET
• SAC/NO   UNDER   LOSS
• MEM/BKN   OVER   LOSS
• PHI/NY   OVER   WIN
• SAN/PHX   UNDER   WIN
• BOS/POR   NO BET
• CHI/OKC   NO BET

FEB 25, 2013

• WAS/TOR   NO BET
• ATL/DET   NO BET
• LAL/DEN   UNDER   LOSS
• BOS/UTA   OVER   WIN

did you bet under or over for the lal/den game. coz here's how i see it.

both teams had a road and home scoring average of 211. line was set at 217. to the average bettor, it would seem that he got value to his under bet because of the 6 point margin from the actual line. however it must have been set by vegas that high for a reason. it is bound to go over.

52% would not generate a profit

@TsikenBreast - One of the model's assumptions is that the lines set by Vegas do not necessarily have to be "statistically balanced." Also, the model tests whether or not the margin from the LINE is statistically significant. There are variables that we can change to make the rules for betting more strict or lax, depending on our risk personality. For example, if the risk aversion rate is set to 53% and the skew adjustment is set to -5 points scored per unit of skew, then LAL/DEN would have been a NO BET.

@Frank the Bank - Assuming a 10% cut, 52% will not even break even. You need 52.38% for that. However, 53% would theoretically generate a profit, albeit a small one.

@robe983 - My background is in finance, and I believe sports betting can more consistent than investing in the stock market.

Alrighty, a new day with fresh NBA fixtures. Since this will mark the first time the model will make recommendations BEFORE the games start (my original post was tested a posteriori), I will make final adjustments to the model that will stay consistent for the remainder of the season. The risk aversion rate will remain the same at 53% (my curiosity is to at least break even), and the skew adjustment will be changed from -3 to -10 points per unit of skew. The latter change more heavily emphasizes teams with skewed normal distributions.

FEB 26, 2013 Picks (Track Record 0-0)
Lines as of 12:33PM

ORL/PHI   OVER   188.0
GS/IND   NO BET   197.5
SAC/MIA   UNDER   210.5
CLE/CHI   NO BET   187.5
BKN/NO   OVER   185.5
MIL/DAL   UNDER   211.0
MIN/PHX   NO BET   195.5
CHA/LAC   NO BET   198.5

Your system is to damn complex keep it simple man

@basesloaded101 - Relatively speaking, my system is very simple. There's just a lot of number crunching to do, which makes it seem complicated. The only statistical tools I use are averages, standard deviations, and skews.

Damn, I was honestly thinking exact opposite on all of the red plays in post #8 . The only one I have laid down was warriors and pacers over. Why do you have them as no bet ?

@spazbite - Think of the all the possible scores as a bell curve. The more likely results lie in the middle. For the sake of example, let's assume Team A and Team B play. LINE is given to be 200. However, the model predicts a total of 199. Is 199 different enough from 200 to warrant an UNDER bet? How about if the model prediction was 150? Is 150 different enough from 200 to make a move? The model tests whether or not this variation is statistically significant. Hence, you will see NO BET when the fixture looks to be a coin flip in the eyes of statistics.

6:52PM Update
CLE/CHI OVER 185.5

Good stuff

Interesting tx for sharing

Results (Overall 3-2)

FEB 26, 2013 (3-2)

ORL/PHI   OVER 185.5   LOSS
GS/IND   NO BET
SAC/MIA   UNDER   209.0   LOSS
CLE/CHI   OVER 185.5   WIN
BKN/NO   OVER 185.0   WIN
MIL/DAL   UNDER 212.5   WIN
MIN/PHX   NO BET
CHA/LAC NO BET

first 

I've updated the statistics for Wednesday. Here are the conditions for the games of FEB 27th. I will be providing these ranges in the case I am not around a computer when the fixtures are about to start.

SAC/ORL   OVER >= 185.5   UNDER <= 200.5
TOR/CLE   OVER >= 188   UNDER <= 199.0
DET/WAS   OVER >= 181.5   UNDER <= 198.0
GS/NY   OVER >= 178.0   UNDER <= 206.0
DAL/MEM   OVER >= 187.0   UNDER <= 197.5
MIL/HOU   OVER >= 195.0   UNDER <= 210.5
NO/OKC   OVER >= 201.5   UNDER <= 207.0
PHX/SAN   OVER >= 189.0   UNDER <= 198.0
ATL/UTA   OVER >= 188.0   UNDER <= 205.5
DEN/POR   OVER >= 201.0   UNDER <= 211.0

PLEASE NOTE:

My mistake, the <> signs should be switched.

For example:
SAC/ORL   OVER <= 185.5   UNDER >=200.5

means, OVER recommendation if LINE is less than or equal to 185.5
UNDER recommendation if LINE is greater than or equal to 200.5
NO BET otherwise

Interesting results so far...I had similar over calculation with Bulls game the only game I played today...comparing my calculations with yours made me wager on it...tx for your system much appreciated sir!

interesting... hope this works out well for you. took me a while to understand this but finally got a grip of it...nice work

first last is sports mavin

Just wanted to check in and make sure I'm reading this correctly..of the Over/Under that was documented. There are 4 Plays tonight: